PanelTools

Collection of (hopefully) useful functions for dealing with panel datasets.

Public functions

pcum(const series y)

Compute cumulated value for each panel unit. This is an alternative to gretl's built-in cum() function which also computes cumulated values for each cross-sectional unit but handles NAs differently (see below).

Parameters

• y: series, target series

Returns

Series with cumulated sum for each cross-sectional unit. NAs are replaced by zero before cumulation for each unit.

Note: In contrast, gretl's built-in cum() function stops computing cumulated values for a unit if the series has NAs between observations.

pmode(const series y)

Compute most frequent value for each cross-sectional unit.

Parameters

• y: series, target series

Returns

Series with most frequent values for each cross-sectional unit. NA values are ignored. If all values are NA for a unit, a series consisting only NA values is returned for that unit.

pfirst(const series y)

Compute first valid value for each panel unit.

Parameters

• y: series, target series

Returns

Series with the first valid value of y for each cross-sectional unit. If no valid values exists for a unit, NA is returned.

plast(const series y)

Compute last valid value for each panel unit.

Parameters

• y: series, target series

Returns

Series with the last valid value of y for each cross-sectional unit. If no valid values exists for a unit, NA is returned.

pquantile(const series y, const scalar value "p-th quantile")

Compute the p-th quantile for each unit.

Parameters

• y: series, target series
• value: scalar, p-th quantile; must be in the range 0 < value < 1

Returns

Series with the p-th quantile for each cross-sectional unit. NA values are ignored. If all values are NA for a unit, a series consisting only NA values is returned for that unit.

pxquantile(const series y, const scalar value "p-th quantile")

Compute the p-th quantile for each time period across all units.

Parameters

• y: series, target series
• value: scalar, p-th quantile; must be in the range 0 < value < 1

Returns

Series with the p-th quantile for each time period across-sectional units. NA values are ignored. If all values are NA for a unit, a series consisting only NA values is returned for that unit.

pxmean(const series y)

Compute mean value of y across all units per time period. Missing values are ignored for computing the mean.

Parameters

• y, series, target series

Returns

Series with the mean value value across all units per observation. NA values are completely ignored and are not counted.

pxfraction(const series y, const scalar value "Value to search",
           const bool fraction[FALSE] "Fraction instead of total number")

Compute the total number of observations of "value" in y across all units per time period. Missing values are completely ignored.

Parameters

• y: series, target series
• value: scalar, Value to search in y
• fraction: bool, if TRUE (=1) compute the fraction of observations taking the value value, if FALSE (=0, default) compute the number of observations instead.

Returns

Series with the respective value across all units per observation. NA values are completely ignored and are not counted.

ppolyfit(const series y, const int order[1::])

Fits a polynomial trend of order order to the input series y using OLS for each cross-sectional unit separately. This function mimics gretl's built-in function polyfit() which works only for time-series data correctly before version 2021c.

Parameters

• y: series, target series
• order: integer, Order of the polynomial trend

Returns

Series of fitted values for each unit. In case y has missing values for a specific unit, for this unit a time-series with NAs is returned.

panelinfo()

Compute basic information about the panel dimension.

Returns

Nothing.

mpmean(const matrix m, const int N)

Compute vector of mean values of m for each cross-sectional unit. Matrix equivalent to gretl's built-in pmean() function. Missing values are ignored.

Parameters

• m: matrix, Column vector of dimension TxN
• N: int, Number of cross-sectional units

Returns

Matrix of stacked time-series of means of m for each unit. NA values are completely ignored.

fe_estimator (const matrix y, const matrix X, const int N[1::])

Matrix-based equivalent of gretl's built-in fixed-effect panel estimator (see help panel). Currently no inference is done.

Parameters

• y: matrix, Column vector of stacked time-series
• X: matrix, Matrix of regressors of stacked time-series
• N: integer, Number of cross-sectional units

Returns

Bundle object comprising both point estimates of coefficients (key: coeff) and estimated fixed-effects stored as stacked time-series (key: ahat).

pols_estimator (const matrix y, const matrix X, const int N[1::])

Matrix-based equivalent of gretl's built-in pooled ols command (see help panel).

Parameters

• y: matrix, Column vector of stacked time-series
• X: matrix, Matrix of regressors of stacked time-series
• N: integer, Number of cross-sectional units

Returns

Bundle object comprising point estimates of coefficients (key: coeff).

pfcast (const matrix X, const bundle Model, const bool verbose[FALSE])

Compute static predictions for a given model. For the fixed-effects, the fixed-effects are added. This is the same treatment as gretl does for the panel fixed-effects model when applying the fcast command.

Parameters

• X: matrix, Matrix of regressors of stacked time-series for test set.
• Model: bundle, Model object holding information on estimation results
• verbose: bool, print details (optional, default = FALSE)

Returns

Matrix of stacked forecasts. N stacked time-series of length horizon. The horizon is internally determined by the input dimension of X and the number of cross-sectional units N as specified in the Model bundle.

mean_robust (const matrix X)

Works as gretl's built-in function meanc(). Compute mean values for each column even in case of missing values.

Parameters

• X: matrix, Matrix with data and possibly missing values

Returns

Row vector with mean values for each column.

sdc_robust (const matrix X)

Works as gretl's built-in function sdc(). Compute standard deviation for each column even in case of missing values.

Parameters

• X: matrix, Matrix with data and possibly missing values

Returns

Row vector with standard deviation values for each column.

panplot_quantile(const series y,
                 const bool do_pxquantile[FALSE],
                 bundle self[null])

Plot quantiles over time.

Parameters

• y: series, target series
• do_pxquantile: bool, If FALSE (default) compute the quantiles across all observations. If TRUE compute the quantiles for each time period across all units.
• self: bundle, Optionally pass parameters for tweaking the plot.

Supported parameters for self are:

• dateformat: string (default "%Y-%m") Set date format of xtics. Requires that panel-time is set via e.g. setobs 4 "2010:1" --panel-time
• fontsize: scalar (default 10)
• fontsize_tics: scalar (default 8)
• filename: string (default "display", immediately show on screen)
• key_position: string (default "top left")
• noenhanced: string (default "noenhanced")
• offset_xtics_x: scalar (default = 0) Offset xtics along the x-axis
• offset_xtics_y: scalar (default = 0) Offset xtics along the y-axis
• options_string: string (default ""; gnuplot strings e.g. set logscale y 2)
• rotate_xtics: scalar (default = 0) Rotate xtics by some degrees
• title: string (default "")
• xlabel: string (default "")
• ylabel: string (default "")
• yrange_min: scalar (default NA (automatically set))
• yrange_max: scalar (default NA (automatically set))
• quantiles: matrix (default {0.05, 0.5, 0.95})

Returns

FALSE if no error occurs, otherwise TRUE.

Changelog

• v0.6 (April 2023)

  • Print dates on x-axis for the panplot_quantile() function when the accessor $obsdate is available (panel time must be set)
  • Make use of Markdown format for help text

• v0.5 (March 2023)

  • panplot_quantile() now returns an error value
  • Add new optional parameter noenhanced for panplot_quantile() for deciding whether to consider underscores in labels and titles (default) or not
  • Bugfix: Parameter options_string did not work before for panplot_quantile()

• v0.4 (October 2022)

  • add new pxquantile() function
  • add new panplot_quantile() function

• v0.3 (March 2022)

  • Add new functions mpmean(), meanc_robust() and sdc_robust()
  • Add new function fe_estimator() for estimating a fixed-effects model using matrices as inputs (no inference supported, yet)
  • Add new function pols_estimator() for estimating a pooled-OLS model using matrices as inputs
  • Add new function pfcast() for computing predictions using some supported panel model using matrices.

• v0.2 (July 2021)

  • Add new ppolyfit() function

• v0.1 (July 2021)

  • Initial version